2 * Code for replacing ftrace calls with jumps.
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
6 * Thanks goes out to P.A. Semi, Inc for supplying me with a PPC64 box.
8 * Added function graph tracer code, taken from x86 that was written
9 * by Frederic Weisbecker, and ported to PPC by Steven Rostedt.
13 #include <linux/spinlock.h>
14 #include <linux/hardirq.h>
15 #include <linux/uaccess.h>
16 #include <linux/module.h>
17 #include <linux/ftrace.h>
18 #include <linux/percpu.h>
19 #include <linux/init.h>
20 #include <linux/list.h>
22 #include <asm/cacheflush.h>
23 #include <asm/code-patching.h>
24 #include <asm/ftrace.h>
25 #include <asm/syscall.h>
28 #ifdef CONFIG_DYNAMIC_FTRACE
30 ftrace_call_replace(unsigned long ip, unsigned long addr, int link)
34 addr = ppc_function_entry((void *)addr);
36 /* if (link) set op to 'bl' else 'b' */
37 op = create_branch((unsigned int *)ip, addr, link ? 1 : 0);
43 ftrace_modify_code(unsigned long ip, unsigned int old, unsigned int new)
45 unsigned int replaced;
48 * Note: Due to modules and __init, code can
49 * disappear and change, we need to protect against faulting
50 * as well as code changing. We do this by using the
51 * probe_kernel_* functions.
53 * No real locking needed, this code is run through
54 * kstop_machine, or before SMP starts.
57 /* read the text we want to modify */
58 if (probe_kernel_read(&replaced, (void *)ip, MCOUNT_INSN_SIZE))
61 /* Make sure it is what we expect it to be */
65 /* replace the text with the new text */
66 if (patch_instruction((unsigned int *)ip, new))
73 * Helper functions that are the same for both PPC64 and PPC32.
75 static int test_24bit_addr(unsigned long ip, unsigned long addr)
77 addr = ppc_function_entry((void *)addr);
79 /* use the create_branch to verify that this offset can be branched */
80 return create_branch((unsigned int *)ip, addr, 0);
85 static int is_bl_op(unsigned int op)
87 return (op & 0xfc000003) == 0x48000001;
90 static unsigned long find_bl_target(unsigned long ip, unsigned int op)
94 offset = (op & 0x03fffffc);
96 if (offset & 0x02000000)
99 return ip + (long)offset;
104 __ftrace_make_nop(struct module *mod,
105 struct dyn_ftrace *rec, unsigned long addr)
108 unsigned long entry, ptr;
109 unsigned long ip = rec->ip;
112 /* read where this goes */
113 if (probe_kernel_read(&op, (void *)ip, sizeof(int)))
116 /* Make sure that that this is still a 24bit jump */
118 printk(KERN_ERR "Not expected bl: opcode is %x\n", op);
122 /* lets find where the pointer goes */
123 tramp = (void *)find_bl_target(ip, op);
125 pr_devel("ip:%lx jumps to %p", ip, tramp);
127 if (!is_module_trampoline(tramp)) {
128 printk(KERN_ERR "Not a trampoline\n");
132 if (module_trampoline_target(mod, tramp, &ptr)) {
133 printk(KERN_ERR "Failed to get trampoline target\n");
137 pr_devel("trampoline target %lx", ptr);
139 entry = ppc_global_function_entry((void *)addr);
140 /* This should match what was called */
142 printk(KERN_ERR "addr %lx does not match expected %lx\n",
148 * Our original call site looks like:
153 * Milton Miller pointed out that we can not simply nop the branch.
154 * If a task was preempted when calling a trace function, the nops
155 * will remove the way to restore the TOC in r2 and the r2 TOC will
158 * Use a b +8 to jump over the load.
160 op = 0x48000008; /* b +8 */
162 if (patch_instruction((unsigned int *)ip, op))
170 __ftrace_make_nop(struct module *mod,
171 struct dyn_ftrace *rec, unsigned long addr)
175 unsigned long ip = rec->ip;
178 if (probe_kernel_read(&op, (void *)ip, MCOUNT_INSN_SIZE))
181 /* Make sure that that this is still a 24bit jump */
183 printk(KERN_ERR "Not expected bl: opcode is %x\n", op);
187 /* lets find where the pointer goes */
188 tramp = find_bl_target(ip, op);
191 * On PPC32 the trampoline looks like:
192 * 0x3d, 0x80, 0x00, 0x00 lis r12,sym@ha
193 * 0x39, 0x8c, 0x00, 0x00 addi r12,r12,sym@l
194 * 0x7d, 0x89, 0x03, 0xa6 mtctr r12
195 * 0x4e, 0x80, 0x04, 0x20 bctr
198 pr_devel("ip:%lx jumps to %lx", ip, tramp);
200 /* Find where the trampoline jumps to */
201 if (probe_kernel_read(jmp, (void *)tramp, sizeof(jmp))) {
202 printk(KERN_ERR "Failed to read %lx\n", tramp);
206 pr_devel(" %08x %08x ", jmp[0], jmp[1]);
208 /* verify that this is what we expect it to be */
209 if (((jmp[0] & 0xffff0000) != 0x3d800000) ||
210 ((jmp[1] & 0xffff0000) != 0x398c0000) ||
211 (jmp[2] != 0x7d8903a6) ||
212 (jmp[3] != 0x4e800420)) {
213 printk(KERN_ERR "Not a trampoline\n");
217 tramp = (jmp[1] & 0xffff) |
218 ((jmp[0] & 0xffff) << 16);
222 pr_devel(" %lx ", tramp);
226 "Trampoline location %08lx does not match addr\n",
233 if (patch_instruction((unsigned int *)ip, op))
239 #endif /* CONFIG_MODULES */
241 int ftrace_make_nop(struct module *mod,
242 struct dyn_ftrace *rec, unsigned long addr)
244 unsigned long ip = rec->ip;
245 unsigned int old, new;
248 * If the calling address is more that 24 bits away,
249 * then we had to use a trampoline to make the call.
250 * Otherwise just update the call site.
252 if (test_24bit_addr(ip, addr)) {
254 old = ftrace_call_replace(ip, addr, 1);
256 return ftrace_modify_code(ip, old, new);
259 #ifdef CONFIG_MODULES
261 * Out of range jumps are called from modules.
262 * We should either already have a pointer to the module
263 * or it has been passed in.
265 if (!rec->arch.mod) {
267 printk(KERN_ERR "No module loaded addr=%lx\n",
273 if (mod != rec->arch.mod) {
275 "Record mod %p not equal to passed in mod %p\n",
279 /* nothing to do if mod == rec->arch.mod */
283 return __ftrace_make_nop(mod, rec, addr);
285 /* We should not get here without modules */
287 #endif /* CONFIG_MODULES */
290 #ifdef CONFIG_MODULES
293 __ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
296 void *ip = (void *)rec->ip;
298 /* read where this goes */
299 if (probe_kernel_read(op, ip, sizeof(op)))
308 * The load offset is different depending on the ABI. For simplicity
309 * just mask it out when doing the compare.
311 if ((op[0] != 0x48000008) || ((op[1] & 0xffff0000) != 0xe8410000)) {
312 printk(KERN_ERR "Unexpected call sequence: %x %x\n",
317 /* If we never set up a trampoline to ftrace_caller, then bail */
318 if (!rec->arch.mod->arch.tramp) {
319 printk(KERN_ERR "No ftrace trampoline\n");
323 /* Ensure branch is within 24 bits */
324 if (!create_branch(ip, rec->arch.mod->arch.tramp, BRANCH_SET_LINK)) {
325 printk(KERN_ERR "Branch out of range");
329 if (patch_branch(ip, rec->arch.mod->arch.tramp, BRANCH_SET_LINK)) {
330 printk(KERN_ERR "REL24 out of range!\n");
338 __ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
341 unsigned long ip = rec->ip;
343 /* read where this goes */
344 if (probe_kernel_read(&op, (void *)ip, MCOUNT_INSN_SIZE))
347 /* It should be pointing to a nop */
348 if (op != PPC_INST_NOP) {
349 printk(KERN_ERR "Expected NOP but have %x\n", op);
353 /* If we never set up a trampoline to ftrace_caller, then bail */
354 if (!rec->arch.mod->arch.tramp) {
355 printk(KERN_ERR "No ftrace trampoline\n");
359 /* create the branch to the trampoline */
360 op = create_branch((unsigned int *)ip,
361 rec->arch.mod->arch.tramp, BRANCH_SET_LINK);
363 printk(KERN_ERR "REL24 out of range!\n");
367 pr_devel("write to %lx\n", rec->ip);
369 if (patch_instruction((unsigned int *)ip, op))
374 #endif /* CONFIG_PPC64 */
375 #endif /* CONFIG_MODULES */
377 int ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
379 unsigned long ip = rec->ip;
380 unsigned int old, new;
383 * If the calling address is more that 24 bits away,
384 * then we had to use a trampoline to make the call.
385 * Otherwise just update the call site.
387 if (test_24bit_addr(ip, addr)) {
390 new = ftrace_call_replace(ip, addr, 1);
391 return ftrace_modify_code(ip, old, new);
394 #ifdef CONFIG_MODULES
396 * Out of range jumps are called from modules.
397 * Being that we are converting from nop, it had better
398 * already have a module defined.
400 if (!rec->arch.mod) {
401 printk(KERN_ERR "No module loaded\n");
405 return __ftrace_make_call(rec, addr);
407 /* We should not get here without modules */
409 #endif /* CONFIG_MODULES */
412 int ftrace_update_ftrace_func(ftrace_func_t func)
414 unsigned long ip = (unsigned long)(&ftrace_call);
415 unsigned int old, new;
418 old = *(unsigned int *)&ftrace_call;
419 new = ftrace_call_replace(ip, (unsigned long)func, 1);
420 ret = ftrace_modify_code(ip, old, new);
425 static int __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
427 unsigned long ftrace_addr = (unsigned long)FTRACE_ADDR;
430 ret = ftrace_update_record(rec, enable);
433 case FTRACE_UPDATE_IGNORE:
435 case FTRACE_UPDATE_MAKE_CALL:
436 return ftrace_make_call(rec, ftrace_addr);
437 case FTRACE_UPDATE_MAKE_NOP:
438 return ftrace_make_nop(NULL, rec, ftrace_addr);
444 void ftrace_replace_code(int enable)
446 struct ftrace_rec_iter *iter;
447 struct dyn_ftrace *rec;
450 for (iter = ftrace_rec_iter_start(); iter;
451 iter = ftrace_rec_iter_next(iter)) {
452 rec = ftrace_rec_iter_record(iter);
453 ret = __ftrace_replace_code(rec, enable);
455 ftrace_bug(ret, rec->ip);
461 void arch_ftrace_update_code(int command)
463 if (command & FTRACE_UPDATE_CALLS)
464 ftrace_replace_code(1);
465 else if (command & FTRACE_DISABLE_CALLS)
466 ftrace_replace_code(0);
468 if (command & FTRACE_UPDATE_TRACE_FUNC)
469 ftrace_update_ftrace_func(ftrace_trace_function);
471 if (command & FTRACE_START_FUNC_RET)
472 ftrace_enable_ftrace_graph_caller();
473 else if (command & FTRACE_STOP_FUNC_RET)
474 ftrace_disable_ftrace_graph_caller();
477 int __init ftrace_dyn_arch_init(void)
481 #endif /* CONFIG_DYNAMIC_FTRACE */
483 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
485 #ifdef CONFIG_DYNAMIC_FTRACE
486 extern void ftrace_graph_call(void);
487 extern void ftrace_graph_stub(void);
489 int ftrace_enable_ftrace_graph_caller(void)
491 unsigned long ip = (unsigned long)(&ftrace_graph_call);
492 unsigned long addr = (unsigned long)(&ftrace_graph_caller);
493 unsigned long stub = (unsigned long)(&ftrace_graph_stub);
494 unsigned int old, new;
496 old = ftrace_call_replace(ip, stub, 0);
497 new = ftrace_call_replace(ip, addr, 0);
499 return ftrace_modify_code(ip, old, new);
502 int ftrace_disable_ftrace_graph_caller(void)
504 unsigned long ip = (unsigned long)(&ftrace_graph_call);
505 unsigned long addr = (unsigned long)(&ftrace_graph_caller);
506 unsigned long stub = (unsigned long)(&ftrace_graph_stub);
507 unsigned int old, new;
509 old = ftrace_call_replace(ip, addr, 0);
510 new = ftrace_call_replace(ip, stub, 0);
512 return ftrace_modify_code(ip, old, new);
514 #endif /* CONFIG_DYNAMIC_FTRACE */
517 extern void mod_return_to_handler(void);
521 * Hook the return address and push it in the stack of return addrs
522 * in current thread info.
524 void prepare_ftrace_return(unsigned long *parent, unsigned long self_addr)
528 struct ftrace_graph_ent trace;
529 unsigned long return_hooker = (unsigned long)&return_to_handler;
531 if (unlikely(atomic_read(¤t->tracing_graph_pause)))
535 /* non core kernel code needs to save and restore the TOC */
536 if (REGION_ID(self_addr) != KERNEL_REGION_ID)
537 return_hooker = (unsigned long)&mod_return_to_handler;
540 return_hooker = ppc_function_entry((void *)return_hooker);
543 * Protect against fault, even if it shouldn't
544 * happen. This tool is too much intrusive to
545 * ignore such a protection.
548 "1: " PPC_LL "%[old], 0(%[parent])\n"
549 "2: " PPC_STL "%[return_hooker], 0(%[parent])\n"
550 " li %[faulted], 0\n"
553 ".section .fixup, \"ax\"\n"
554 "4: li %[faulted], 1\n"
558 ".section __ex_table,\"a\"\n"
564 : [old] "=&r" (old), [faulted] "=r" (faulted)
565 : [parent] "r" (parent), [return_hooker] "r" (return_hooker)
569 if (unlikely(faulted)) {
575 trace.func = self_addr;
576 trace.depth = current->curr_ret_stack + 1;
578 /* Only trace if the calling function expects to */
579 if (!ftrace_graph_entry(&trace)) {
584 if (ftrace_push_return_trace(old, self_addr, &trace.depth, 0) == -EBUSY)
587 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
589 #if defined(CONFIG_FTRACE_SYSCALLS) && defined(CONFIG_PPC64)
590 unsigned long __init arch_syscall_addr(int nr)
592 return sys_call_table[nr*2];
594 #endif /* CONFIG_FTRACE_SYSCALLS && CONFIG_PPC64 */